Fj?0UdZddlZddlZddlZddlZddlZddlZddlmZddl m Z m Z m Z ddl mZddlmZddlmZddlmZmZmZejeZdqd ZddlmZd ed efd Z e d Z!e dZ"e dZ#dZ$dZ%dZ&ej'ddZ(ej'ddZ)ej'ddZ*dZ+dZ,dZ-ej'ddZ.ej'ddZ/ej'd d!Z0hd"Z1hd#Z2d$Z3hd%Z4hd&Z5da6e e7e8d'<da9e ee8d(<d e:fd)Z;dqd*e e:d efd+Zd e efd/Z?d e efd0Z@d e efd1ZAd efd2ZBd3e ed efd4ZCd3e ed efd5ZDd efd6ZEeFhd7ZGd8ZHdZId9ZJeFhd:ZKd*e ee fd;ed e ee ffd<ZLd*e ee fd;ed e ee ffd=ZMd>e ee fd efd?ZNd@ed*e ee fd e e ee ffdAZOd*e ee ffdBZPdqd*e e ee fd efdCZQd>e ee fd eRfdDZSd>e ee fd efdEZTdFedGeUd e efdHZVdIedJe ed efdKZWdFedLe eefd efdMZXdNejYd dfdOZZdPedQeRd ej[fdRZ\dSedTedUed efdVZ] dqdWedXed>e ee fd*e ee fdYe ed e ee ff dZZ^d*e:d efd[Z_ dqddd\dWed@ed*e e ee fd]e ed^e ed e e ee ff d_Z`dWed e e ee ffd`ZadaZbdbecd efdcZdd3efddZedWed3ed e ee ffdeZfdWedfed ege ee effdgZhdWed3ed e ee ffdhZidWed3ed e ee ffdiZjdWed3ed e ee ffdjZkdWed3ed e ee ffdkZldWed3ed e ee ffdlZmdqdWed]e ed e ee ffdmZnd egeeffdnZodoe d efdpZpdS)ru Transcription Tools Module Provides speech-to-text transcription with six providers: - **local** (default, free) — faster-whisper running locally, no API key needed. Auto-downloads the model (~150 MB for ``base``) on first use. - **groq** (free tier) — Groq Whisper API, requires ``GROQ_API_KEY``. - **openai** (paid) — OpenAI Whisper API, requires ``VOICE_TOOLS_OPENAI_KEY``. - **mistral** — Mistral Voxtral Transcribe API, requires ``MISTRAL_API_KEY``. - **xai** — xAI Grok STT API, requires ``XAI_API_KEY``. High accuracy, Inverse Text Normalization, diarization, 21 languages. Used by the messaging gateway to automatically transcribe voice messages sent by users on Telegram, Discord, WhatsApp, Slack, and Signal. Supported input formats: mp3, mp4, mpeg, mpga, m4a, wav, webm, ogg, aac Usage:: from tools.transcription_tools import transcribe_audio result = transcribe_audio("/path/to/audio.ogg") if result["success"]: print(result["transcript"]) N)Path)OptionalDictAny)urljoin)is_truthy_value)resolve_managed_tool_gateway)managed_nous_tools_enabled%nous_tool_gateway_unavailable_messageresolve_openai_audio_api_keyc~ ddlm}n%#t$rtj||cYSwxYw||}||n|S)aRead env values through the live config module. Tests may monkeypatch and later restore ``hermes_cli.config.get_env_value`` before this module is imported. Resolve the helper at call time so STT does not keep a stale imported function for the rest of the test process. r) get_env_value)hermes_cli.configr ImportErrorosgetenv)namedefault_get_env_valuevalues 8/usr/local/lib/hermes-agent/tools/transcription_tools.pyrr1sl(EEEEEEE (((yw'''''( N4 Em77.s  ++ module_namereturnc tj|duS#ttf$r&|t vp|t jjvcYSwxYwN)_ilu find_specr ValueErrorglobalsrsysmodules)rs r_safe_find_specr"FsdI~k**$66  $IIIgii'H;"&.+HHHHIs4AAfaster_whisperopenai mistralailocalbaseenSTT_OPENAI_MODEL whisper-1STT_GROQ_MODELwhisper-large-v3-turboSTT_MISTRAL_MODELzvoxtral-mini-latestHERMES_LOCAL_STT_COMMANDHERMES_LOCAL_STT_LANGUAGE)z/opt/homebrew/binz/usr/local/bin GROQ_BASE_URLzhttps://api.groq.com/openai/v1STT_OPENAI_BASE_URLzhttps://api.openai.com/v1XAI_STT_BASE_URLzhttps://api.x.ai/v1> .aac.m4a.mp3.mp4.ogg.flac.mpeg.mpga.webm.wav>.aif.aiffr<i>gpt-4o-transcribegpt-4o-mini-transcriber*>whisper-large-v3distil-whisper-large-v3-enr, _local_model_local_model_namecp ddlm}|diS#t$ricYSwxYw)zDLoad the ``stt`` section from user config, falling back to defaults.r load_configstt)rrGget ExceptionrFs r_load_stt_configrKusY111111{}}  +++  s #& 55 stt_configcp|t}|dd}t|dS)z(Return whether STT is enabled in config.NenabledT)r)rKrIr)rLrNs ris_stt_enabledrO~s9%'' nnY--G 7D 1 1 11cF tdS#t$rYdSwxYw)zbReturn True when OpenAI audio can use config credentials, env credentials, or the managed gateway.TF)#_resolve_openai_audio_client_configrrPr_has_openai_audio_backendrTs:+---t uus    binary_namectD]X}t||z }|r0tj|tjrt |cSYtj|S)zMFind a local binary, checking common Homebrew/local prefixes as well as PATH.) COMMON_LOCAL_BIN_DIRSrexistsraccessX_OKstrshutilwhich)rU directory candidates r _find_binaryr`sm*"" OOk1      "")Irw"?"? "y>> ! ! ! < $ $$rPc tdS)Nffmpegr`rSrPr_find_ffmpeg_binaryrds  ! !!rPc tdS)NwhisperrcrSrPr_find_whisper_binaryrgs  " ""rPctjtd}|r|St }|rt j|}|dSdS)Nza {input_path} --model {model} --output_format txt --output_dir {output_dir} --language {language})rrLOCAL_STT_COMMAND_ENVstriprgshlexquote) configuredwhisper_binary quoted_binarys r_get_local_command_templaterqsn0"55;;==J)++N  N33  > > >  4rPc"tduSr)rqrSrPr_has_local_commandrss & ( ( 44rP model_namec|r|tvs |tvr<|r3|tvs |tvr!td|ttS|S)aReturn a valid faster-whisper model size, mapping cloud-only names to the default. Cloud providers like OpenAI use names such as ``whisper-1`` which are not valid for faster-whisper (which expects ``tiny``, ``base``, ``small``, ``medium``, or ``large-v*``). When such a name is detected we fall back to the default local model and emit a warning so the user knows what happened. zSTT model '%s' is a cloud-only name and cannot be used with the local provider. Falling back to '%s'. Set stt.local.model to a valid faster-whisper size (tiny, base, small, medium, large-v3).) OPENAI_MODELS GROQ_MODELSloggerwarningDEFAULT_LOCAL_MODELrts r_normalize_local_modelr|sn  #}44 k8Q8Q  :66* :S:S NNM#    #" rPc t|Sr)r|r{s r_normalize_local_command_modelr~s !* - --rPc ddlm}|dddlm}|drdSn2#t $r%}t d|Yd}~nd}~wwxYwdS) aoAttempt to lazy-install faster-whisper and return True on success. The module-level ``_HAS_FASTER_WHISPER`` flag is set at import time and cached. If the package wasn't installed at startup, calling ``ensure()`` installs it. This function re-checks dynamically after installation so the provider can use it immediately without a process restart. r)ensurezstt.faster_whisperNr#Tz)Lazy install of faster-whisper failed: %sF)tools.lazy_depsrimportlib.utilutilrrJrxdebug)r_iuexcs r_try_lazy_install_sttrsG******#$$$$$$$$$ ==) * * 4  GGG @#FFFFFFFFG 5s,1 A AA >xaigroqr&r$mistral local_command,txt>srtrvttjsonrct|tsiS||}t|tr|niS)z=Return an stt sub-section if it's a dict, else an empty dict.) isinstancedictrI)rLrsections r_get_stt_sectionr sC j$ ' ' nnT""G $// 777R7rPc t|d}t|tr||nd}t|tr|S|t vrt||}|r|SiS)uReturn the config dict for a user-declared STT command provider. Looks up ``stt.providers.`` first (the canonical location), and falls back to ``stt.`` so users who followed the built-in layout still work. Returns an empty dict when the provider is not declared. Built-in names are NOT special-cased here — the caller short-circuits them before this is consulted, AND ``_is_command_stt_provider_config`` requires an explicit ``command:`` value, so a built-in section like ``stt.openai`` (which has ``model``/``language`` but no ``command``) can't accidentally be treated as a command provider. providersN)rrrrIlowerBUILTIN_STT_PROVIDERS)rLrrrlegacys r_get_named_stt_provider_configrs ![99I%/ 4%@%@JimmD!!!dG'4   zz||000!*d33  M IrPconfigcjt|tsdSt|dpd}|r|dkrdS|d}t|to t |S)z?Return True when *config* declares a command-type STT provider.Ftypericommand)rrr[rIrkrbool)rptypers r_is_command_stt_provider_configr0s fd # #u  6""(b ) ) / / 1 1 7 7 9 9E )##ujj##G gs # # =W]]__(=(==rPproviderc|sdS|}|tvs|dkrdSt||}t |r|SdS)zReturn the provider config if *provider* resolves to a command type. Built-in provider names are rejected (they have native handlers). Returns None when the name is a built-in, ``"none"``, unknown, or not a command type. Nnone)rrkrrr)rrLkeyrs r$_resolve_command_stt_provider_configr;sn t ..   " "C ###sf}}t +J < > ttt 5rPc|d|dt} t|}n+#ttf$rttcYSwxYw|dkrttS|S)z5Return timeout in seconds, falling back when invalid.timeouttimeout_secondsr)rI#DEFAULT_COMMAND_STT_TIMEOUT_SECONDSfloat TypeErrorr)rrawrs r_get_command_stt_timeoutrds **Y +<>a b b c cC:c  z ":::899999: zz8999 LsA%A)(A)c|dp|dpt}t|d}|t vr|ntS)z6Return the validated output format (txt/json/srt/vtt).format output_format.)rI!DEFAULT_COMMAND_STT_OUTPUT_FORMATr[rrklstripCOMMAND_STT_OUTPUT_FORMATS)rrfmts r_get_command_stt_output_formatrps{  8 - ::o & & - , c((..   " " ) )# . .C333339ZZrPcommand_templatepositioncd}d}d}||kr\||}|dkr |dkrd}n:|dkr|rd}n/|dkrd}n&|dkrd}n|dkrd}n|dkrd}n |dkr|dz }|dz }||k\|S) uReturn the shell quote character active right before *position*. Mirrors ``tools.tts_tool._shell_quote_context`` — kept local to avoid cross-module import of a private helper. Returns ``"'"`` / ``'"'`` when inside a quoted region, ``None`` for bare context. NFr'"\TrS)rrrmescapedichars r_shell_quote_context_sttr{s EG A h,," C<<s{{ c\\  S[[EE S[[EE T\\ FA Q% h,,& LrPr quote_contextc\|dkr|ddS|dkrR|ddddddd d Stjd krtj|gSt j|S) zQuote a placeholder value for its position in a shell command template. Mirrors ``tools.tts_tool._quote_command_tts_placeholder``. rz'\''rrz\\z\"$z\$`z\`nt)replacerr subprocess list2cmdlinerlrm)rrs r_quote_command_stt_placeholderrs }}S'***  WT6 " " WS% WS% WS%   w$&w/// ;u  rP placeholderscr ddldfdD}d|d|d}g dd d tf fd }||}|d d dd} D]\}}|||}|S)aReplace supported placeholders while preserving ``{{`` / ``}}``. Mirrors ``tools.tts_tool._render_command_tts_template``. Placeholders are shell-quote-aware: ``{voice}`` inside single quotes gets single-quote-safe escaping, inside double quotes gets ``$``/`` ` ``/`` " `` escaping, outside quotes gets ``shlex.quote``. Doubled braces ``{{`` and ``}}`` are preserved as literal ``{`` / ``}`` for users who want to embed JSON snippets in their command. rN|c3BK|]}|VdSr)escape).0rres r z/_render_command_stt_template..s->>RYYt__>>>>>>rPz(?z)\}\}|\{(?Pz)\})matchz re.Match[str]rc |dp|d}dtd}|t|t |f|S)Ndoublesingle__HERMES_STT_PLACEHOLDER___)grouplenappendrrstart)rrtokenrr replacementss r replace_matchz3_render_command_stt_template..replace_matchs{{8$$= H(=(=AC ,=,=AAA  *T"()95;;==II       rPz{{{z}}})rjoincompiler[subr) rrnamespatternrrenderedrrrrs `` @@r_render_command_stt_templatersIII HH>>>>>>> > >EjjPuPPEPPPG+-L _         {{=*:;;Hc**224==H$22 u##E511 OrPprocc |dStjdkrk tjddddt |jgtjtjdn$#t$r| YnwxYwdS d dl }nf#t$rY|  | d n)#tj$r| YnwxYwYdSwxYw ||j}|d D]'} | #|j$rY$wxYw| n0#|j$rYdSt$r| YnwxYw | d dS#tj$rYnwxYw ||j}|d D]'} | #|j$rY$wxYw| dS#|j$rYdSt$r| YdSwxYw)zBest-effort termination of a shell process and all of its children. Mirrors ``tools.tts_tool._terminate_command_tts_process_tree``. Nrtaskkillz/Fz/Tz/PID)stdoutstderrrrrT) recursive)pollrrrrunr[pidDEVNULLrJkillpsutilr terminatewaitTimeoutExpiredProcesschildren NoSuchProcess)rrparentchilds r#_terminate_command_stt_process_treer s  yy{{ w$  NT4TX?!)!)           IIKKKKK      IIaI (    IIKKKKK  ))__t_44  E !!!!'          !   $      ))__t_44  E  '        sAA,,B  B BC;7C C;#C41C;3C44C;:C;?2E,2EE, EE,EE,, F9FFF55GG 2H9>HH9 H H9H  H99 I'I'&I'rrcdtjtjdd}tjdkrt tdd|d<nd|d<tj|fi|} ||\}}n#tj$rz}t| |d \}}n2#t$r%t |d d }t |d d }YnwxYwtj|||| |d }~wwxYw|j rtj |j ||| tj ||j ||S)z~Run a command-provider shell command with process-tree timeout cleanup. Mirrors ``tools.tts_tool._run_command_tts``. T)shellrrtextrCREATE_NEW_PROCESS_GROUPr creationflagsstart_new_sessionrroutputNr)rr) rPIPErrgetattrPopen communicaterr rJ returncodeCalledProcessErrorCompletedProcess)rr popen_kwargsrrrrs r_run_command_sttrs // $$L  w$(/ |dd}YnwxYw|r|S|r(|r|St d|d)uReturn the transcript text from a command-provider invocation. Resolution: 1. If ``output_path`` exists and is non-empty → read it (raw text). 2. Else if ``stdout`` is non-empty → use stdout (lets users write curl-style one-liners that emit transcript to stdout instead of writing a file). 3. Else → raise RuntimeError (no usable output produced). For JSON format, we still return the raw bytes — extracting a ``text`` field is out of scope; users either configure ``format: txt`` or post-process JSON downstream. (Same trade-off as TTS: the runner doesn't try to be clever about output shape.) utf-8encodingr)errorsz-Command STT provider wrote no output file at z and produced no stdout)rX read_textrkUnicodeDecodeError read_bytesdecode RuntimeError)rrrcontents r_read_command_stt_outputr'Bs Y!++W+==CCEEGG! Y Y Y!,,..55gi5PPVVXXGGG Y  N &,,..||~~  "  " " "  s(?ABB file_path provider_namemodel_overridec t|dpd}|s dd|d|ddSt|}|s dd|d|dSt |}t|}|dp|dpt} |p|d pd} tj d |d  5} t| d |z } t| t| t| j |t| t| d} t|| }td|j| t#||}n#t$j$rdd|d|d|dddcYcdddSt$j$r}g}|jr/|d|j|jr/|d|jd|pd}dd|d|d|jd|dcYd}~cdddSd}~wwxYw t5| |jpd|}n7#t6$r*}dd|t|dcYd}~cdddSd}~wwxYw dddn #1swxYwYn$#t8$r}dd|d|d|dcYd}~Sd}~wwxYwtd|j|t;|d||dS)a,Transcribe via a user-declared ``stt.providers.: type: command``. Placeholder grammar: | Placeholder | Substituted with | |-------------------|-----------------------------------------------------------| | ``{input_path}`` | absolute path to the audio file (original location) | | ``{output_path}`` | absolute path the provider should write its transcript to | | ``{output_dir}`` | parent dir of ``{output_path}`` | | ``{format}`` | configured output format (``txt`` / ``json`` / ``srt`` / ``vtt``) | | ``{language}`` | configured language code (default ``en``) | | ``{model}`` | configured model id (empty when not set) | All placeholders are shell-quote-aware (see ``_render_command_stt_template``). Doubled braces ``{{`` and ``}}`` are preserved as literal braces. Returns the standard transcribe-response envelope (``success``, ``transcript``, ``provider``, ``error``). rriFzstt.providers.z.command is not configured)success transcriptrerrorAudio file not found: languagemodelzhermes-cmd-stt--prefixz transcript.) input_pathr output_dirrr0r1z0Transcribing %s via command STT provider '%s'...zSTT command provider 'z' timed out after gsNzstderr: zstdout: z; zno command outputz' exited with code z: z ' failed: z7Transcribed %s via command STT provider '%s' (%d chars)Tr,r-r)r[rIrkr expanduserrXrrDEFAULT_COMMAND_STT_LANGUAGEtempfileTemporaryDirectoryresolverrrxinforrrrrrrrrrr'r%OSErrorr)r(r)rrLr*raudiorrr0r1tmpdirrrrresultr detail_partsdetailtranscript_texts r_transcribe_command_sttrG`s<46::i006B77==??  %OmOOO     OO & & ( (E <<>> %9i99    'v..G26::M : ( >>* % % ( '  7fjj11 7RE?  (0R-0R0R0R S S S6 W]v,,)F})F)FFK!%--//22";//!+"455'MMU L33C\RRG KKB M    )'7;;,   $"$ -((("((( %6 6 6 6 6 6 6 6 60   ! :I ''(G3:3C3C3E3E(G(GHHH:I ''(G3:3C3C3E3E(G(GHHH<00G4G$"$ -666>66-366 E6 6 6 6 6 6 6 6 6 " ":!4"m##    $"$ - XX c6 6 6 6 6 6 6 6 `  [6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 p    %LmLLsLL           KKA M3#7#7 %!  s,LB(L .F?>L ?J)L  L+J)9BJ$J)L  L$J))L -KL  K:K5#K:$L ( L5K::L > L LLLL L7 L2,L72L7c"t|sdSd|v}|dt}|r~|dkrEtrdSt rdSt rdSt ddS|dkrOt rdStrt ddSt ddS|dkr4trtd rdSt d dS|d kr3trtrd St d dS|d krt ddS|dkrAddl m }|drdSt ddS|StrdSt rdSt rdStr+td rt ddStr*trt dd S ddl m }|drt ddSn#t$rYnwxYwdS)uDetermine which STT provider to use. When ``stt.provider`` is explicitly set in config, that choice is honoured — no silent cloud fallback. When no provider is configured, auto-detect tries: local > groq (free) > openai (paid). rrr&rzhSTT provider 'local' configured but unavailable (install faster-whisper or set HERMES_LOCAL_STT_COMMAND)z9Local STT command unavailable, using local faster-whisperz7STT provider 'local_command' configured but unavailabler GROQ_API_KEYz7STT provider 'groq' configured but GROQ_API_KEY not setr$z9STT provider 'openai' configured but no API key availableru STT provider 'mistral' (Voxtral Transcribe) is temporarily disabled — `mistralai` PyPI package is quarantined (malicious 2.4.6 release on 2026-05-12). Falling back to another provider. Set stt.provider in config.yaml to 'local' or 'openai' to silence this warning.rrresolve_xai_http_credentialsapi_keyzBSTT provider 'xai' configured but no xAI credentials are availablez.No local STT available, using Groq Whisper APIz0No local STT available, using OpenAI Whisper APIz.No local STT available, using xAI Grok STT API)rOrIDEFAULT_PROVIDER_HAS_FASTER_WHISPERrsrrxryr? _HAS_OPENAIrrTtools.xai_httprKrJ)rLexplicitrrKs r _get_providerrRs/ * % %vZ'H~~j*:;;HB w  " w!## '&$&& w NNK   6  & &!## '&"  WXXXw NNI   6 v   }^<< v NNI   6 x   8:: x NNK   6 y NN7   6 u   C C C C C C++--11)<< u NNT   6 ww}^44 DEEEv022 FGGGx ?????? ' ' ) ) - -i 8 8  KKH I I I5       6s?=I?? J  J r1r0r1r0c|sdS|}|tvs|dkrdS|tt ||rdS ddlm}ddlm}|||}||d||}n3#t$r&} t d| Yd} ~ dSd} ~ wwxYw|dS | } n7#t$r*} t d || d d } Yd} ~ nd} ~ wwxYw| s&td |d d d|d|dStd| ||||} nB#t$r5} t d|| d d d d|d| |dcYd} ~ Sd} ~ wwxYwt!| t"s d d d|d|dS| d|| S)uRoute the call to a plugin-registered transcription provider, or return None. Returns the transcribe-response dict on dispatch, or ``None`` to fall through to the legacy "No STT provider available" error path. Resolution invariants enforced here: 1. Built-in provider names short-circuit — never reach the plugin registry. The caller (``transcribe_audio``) handles ``local``, ``groq``, ``openai``, etc. via its existing elif chain; this function defensively rejects those names so a plugin can't be silently dispatched under a built-in name even if it somehow slipped past the registry's built-in shadow guard. 2. Same-name command-type provider declared under ``stt.providers.: type: command`` wins over a plugin. The caller short-circuits to the command runner before reaching us, but we re-verify here so a refactor of the caller can't silently break the invariant (matches TTS PR #17843 precedence rule). 3. Plugin dispatch fires only when ``provider`` matches a registered :class:`TranscriptionProvider` whose ``name`` equals the configured value. Unknown names with no plugin registered return None (caller surfaces the legacy "No STT provider" message). 4. Availability gating: when the matched plugin reports ``is_available() == False`` (missing API key, missing optional SDK, etc.) this returns an error envelope identifying the plugin as unavailable — **not** ``None`` — because the user explicitly opted into this plugin via ``stt.provider`` and the generic fallthrough message would be misleading. Provider exceptions are caught and converted into the standard error envelope (matches the legacy built-in error shapes — the gateway/CLI caller already expects ``{success: False, error: "...", transcript: ""}`` on failure). Nrr) get_provider)_ensure_plugins_discoveredT)forcez2STT plugin dispatch skipped (discovery failed): %suNSTT plugin provider '%s' is_available() raised: %s — treating as unavailableexc_infoFzRSTT plugin provider '%s' reports not available; returning unavailability envelope.riz STT plugin 'uY' is not available — check that its required credentials / dependencies are configured.)r,r-r.rz-Transcribing with plugin STT provider '%s'...rSz#STT plugin provider '%s' raised: %sz ' raised: z' returned a non-dict resultr)rrkrrragent.transcription_registryrUhermes_cli.pluginsrVrJrxr is_availableryr? transcriberr setdefault) r(rrLr1r0rrUrVplugin_providerr availablerCs r_dispatch_to_plugin_providerraUs#X t ..   " "C ###sf}}t"A&z377##t======AAAAAA""$$$&,s++  " ' &T 2 2 2 2*l3//O  I3OOOtttttt#0022  &'*C$          '(+   FsFFF    KK?EEE  ++ ,        13d    8C88388           fd # # ECEEE     j#&&& MsN:B C #CC C&& D0 DD!E:: F9*F4.F94F9c Tt|}tj|r ddd|dS|s ddd|dS|s ddd|dS|jtvr6ddd|jdd ttdS | j }|tkrddd |d z d d td z dddSn #t$r}ddd|dcYd}~Sd}~wwxYwdS)z>Validate the audio file. Returns an error dict or None if OK.FrizPath is a symbolic link: r,r-r.r/zPath is not a file: zUnsupported format: z . Supported: z, zFile too large: iz.1fzMB (max z.0fzMB)zFailed to access file: N)rrpathislinkrXis_filesuffixrSUPPORTED_FORMATSrsortedstatst_size MAX_FILE_SIZEr@)r( audio_path file_sizees r_validate_audio_filerpsiJ w~~j!!f =dYb=d=deee     c =aV_=a=abbb     a =_T]=_=_```  (999rJ,=rrDIIV\]nVoVoLpLprr   \OO%%- } $ $  uI,CuuuQ^bkQluuuu  % \\\ =ZWX=Z=Z[[[[[[[[\ 4s =D D%D D% D%) libcublaslibcudnn libcudartzcannot be loadedzcannot open shared objectzno kernel image is availablezno CUDA-capable devicez#CUDA driver version is insufficientrcbt|tfdtDS)agHeuristic: is this exception a missing/broken CUDA runtime library? ctranslate2 raises plain RuntimeError with messages like ``Library libcublas.so.12 is not found or cannot be loaded``. We want to catch missing/unloadable shared libs and driver-mismatch errors, NOT legitimate runtime failures ("CUDA out of memory", model bugs, etc.). c3 K|]}|vV dSrrS)rmarkermsgs rrz-_looks_like_cuda_lib_error..)s'CCv}CCCCCCrP)r[any_CUDA_LIB_ERROR_MARKERS)rrws @r_looks_like_cuda_lib_errorrz s4 c((C CCCC+BCCC C CCrPcddlm} ||ddS#t$rC}t|std|||ddcYd}~Sd}~wwxYw) ucLoad faster-whisper with graceful CUDA → CPU fallback. faster-whisper's ``device="auto"`` picks CUDA when the ctranslate2 wheel ships CUDA shared libs, even on hosts where the NVIDIA runtime (``libcublas.so.12`` / ``libcudnn*``) isn't installed — common on WSL2 without CUDA-on-WSL, headless servers, and CPU-only developer machines. On those hosts the load itself sometimes succeeds and the dlopen failure only surfaces at first ``transcribe()`` call. We try ``auto`` first (fast CUDA path when it works), and on any CUDA library load failure fall back to CPU + int8. r WhisperModelautodevice compute_typeufaster-whisper CUDA load failed (%s) — falling back to CPU (int8). Install the NVIDIA CUDA runtime (libcublas/libcudnn) to use GPU.cpuint8N)r#r}rJrzrxry)rtr}rs r_load_local_whisper_modelr,s,+++++ K|JvFKKKK KKK)#..   O    |Ju6JJJJJJJJJKs  A#8AA#A#ctstsddddS t t|kr,td|t |a|atdidptj tpd}d d i}|r||d< tj |fi|\}}d d |D}n#t$r}t|std |dadaddlm}||dda|atj |fi|\}}d d|D}Yd}~nd}~wwxYwtdt'|j||j|jd|ddS#t$r0} td| dddd| dcYd} ~ Sd} ~ wwxYw)z.Transcribe using faster-whisper (local, free).Frizfaster-whisper not installedrcNzELoading faster-whisper model '%s' (first load downloads the model)...r&r0 beam_sizer c3HK|]}|jVdSrr rkrsegments rrz$_transcribe_local..b0!O!O7',"4"4"6"6!O!O!O!O!O!OrPulfaster-whisper CUDA runtime failed mid-transcribe (%s) — evicting cached model and retrying on CPU (int8).rr|rrrc3HK|]}|jVdSrrrs rrz$_transcribe_local..vrrPz;Transcribed %s via local whisper (%s, lang=%s, %.1fs audio)Tr9zLocal transcription failed: %srXLocal transcription failed: )rNrrCrDrxr?rrKrIrrLOCAL_STT_LANGUAGE_ENVr]rrJrzryr#r}rrr0durationr.) r(rt _forced_langtranscribe_kwargssegmentsr?r-rr}ros r_transcribe_localrGs a$&& a$BA_`` `2a  #4 #B#B KK_ak l l l4Z@@L *     " "7B / / 3 3J ? ? y/00   )!,  9,8 j ) P)4YTTBSTTNHd!O!Oh!O!O!OOOJJ P P P .c22  NND     L $  3 3 3 3 3 3'< 5vVVVL * )4YTTBSTTNHd!O!Oh!O!O!OOOJJJJJJ' P*  I OO *dmT]    zwOOO aaa 5q4 HHH =_\]=_=_````````asJBF<74C,+F<, E96A9E4/F<4E99AF<< G6%G1+G61G6work_dirc$t|}|jtvr|dfSt }|sdSt j||jd}|dd||g} tj |ddd|dfS#tj $rn}|j p'|j pt|}t d||dd |fcYd}~Sd}~wwxYw) z.Normalize audio for local CLI STT when needed.N)NzOLocal STT fallback requires ffmpeg for non-WAV inputs, but ffmpeg was not foundr<z-yz-iTcheckcapture_outputr z#ffmpeg conversion failed for %s: %sz'Failed to convert audio for local STT: )rrgrLOCAL_NATIVE_AUDIO_FORMATSrdrrdrstemrrrrrkrr[rxr.)r(rrmrbconverted_pathrrodetailss r_prepare_local_audiors3iJ  $>>>$ " "F gffW\\(z,D,D,DEENtT9n=GIwd4dKKKKt##  (III(..""@ahnn&6&6@#a&& :IwOOOHwHHHHHHHHHIs6BD!A#D D Dc t}|sddtddStdidpt jt pt}t|} tj d5}t||\}}|rdd|dcd d d S| tj|tj|tj|tj| }tt jtd} | rt#j|d d d d n*t#jtj|d d d t)t+|d} | sddddcd d d S| dd} t0dt+|j|t7| d | ddcd d d S#1swxYwYd S#t8$r} dddtd| dcYd } ~ Sd } ~ wt"j$rp} | jp'| jptA| } t0!d|| ddd| dcYd } ~ Sd } ~ wtD$r0} t0!d| d ddd| dcYd } ~ Sd } ~ wwxYw)zNRun the configured local STT command template and read back a .txt transcript.Friz5 not configured and no local whisper binary was foundrcr&r0zhermes-local-stt-r3N)r5r6r0r1T)r rrr rz*.txtzALocal STT command completed but did not produce a .txt transcriptrrrz3Transcribed %s via local STT command (%s, %d chars)rr9zInvalid z template, missing placeholder: z#Local STT command failed for %s: %szLocal STT failed: z7Unexpected error during local command transcription: %srXr)#rqrjrKrIrrrDEFAULT_LOCAL_STT_LANGUAGEr~r<r=rrrlrmrrkrrsplitrirglobr!rxr?rrKeyErrorrrrr[r.rJ)r(rtrr0normalized_modelr6prepared_input prep_errorr use_shell txt_filesrFrors r_transcribe_local_commandrsi244  (___     w++// ;; & 9+ , , & % 6jAA1a  (0C D D D" a )=i)T)T &NJ Q#(ZPP" a" a" a" a" a" a" a" a '-- ;~66 ;z22X..k"233 .GRY'>I $"$`+" a" a" a" a" a" a" a" a6(l44g4FFLLNNO KKEY$ O$$     $?P_``E" a" a" a" a" a" a" a" a" a" a" a" a" a" a" a" a" a" aH    Z 5ZZWXZZ         (]]](..""@ahnn&6&6@#a&& :IwOOO =[RY=[=[\\\\\\\\ aaa NPQ\` aaa =_\]=_=_````````asI-I 4 I-DI  I-$A/I  I- I$$I-'I$(I-- M7J M MA%L M M%M<MMc&td}|sddddStsddddS|tvr(td|t t } dd lm}m}m }m }||td d } t|d 5}|j j||d } dddn #1swxYwYt!| } tdt%|j|t)| d| ddt+|dd} t-| r | SS#t+|dd} t-| r | wwxYw#t.$r ddd|dcYS|$r} ddd| dcYd} ~ Sd} ~ w|$r} ddd| dcYd} ~ Sd} ~ w|$r} ddd| dcYd} ~ Sd} ~ wt0$r0} td| dddd| dcYd} ~ Sd} ~ wwxYw)z8Transcribe using Groq Whisper API (free tier available).rIFrizGROQ_API_KEY not setrcopenai package not installedz(Model %s not available on Groq, using %srOpenAIAPIErrorAPIConnectionErrorAPITimeoutErrorrLbase_urlr max_retriesrbr r1fileresponse_formatNz*Transcribed %s via Groq API (%s, %d chars)Trr9closePermission denied: Connection error: Request timeout: API error: zGroq transcription failed: %srXTranscription failed: )rrOrvrxr?DEFAULT_GROQ_STT_MODELr$rrrrr0openrAtranscriptionscreater[rkrrrrcallablePermissionErrorrJr.) r(rtrLrrrrclient audio_file transcriptionrFrros r_transcribe_groqrshN++G U =STTT ] =[\\\]"" > Lbccc+ [PPPPPPPPPPPP-YZ[[[ i&& * & ; B B$#$*!C!!                "-006688O KKDi-z3;O;O Q Q Q $?PVWWFGT22E  FGT22E   ``` =^S\=^=^_____ WWW =URS=U=UVVVVVVVV VVV =TQR=T=TUUUUUUUU PPP =N1=N=NOOOOOOOO [[[ 4a$ GGG =YVW=Y=YZZZZZZZZ[s F8E#B7+ E7B;;E>B;?A%E$*F-E==FHHF'!H'H/F=7H=HG H H %H H HcT t\}}n*#t$r}ddt|dcYd}~Sd}~wwxYwtsddddS|tvr(t d|tt} ddlm }m }m }m }|||d d } t|d 5} | jj|| |d krd nd} dddn #1swxYwYt#| } t dt%|j|t)| d| ddt+| dd} t-| r | SS#t+| dd} t-| r | wwxYw#t.$r ddd|dcYS|$r}ddd|dcYd}~Sd}~w|$r}ddd|dcYd}~Sd}~w|$r}ddd|dcYd}~Sd}~wt0$r0}t d|dddd|dcYd}~Sd}~wwxYw)z+Transcribe using OpenAI Whisper API (paid).FrircNrz*Model %s not available on OpenAI, using %srrrrrr*r rrz,Transcribed %s via OpenAI API (%s, %d chars)Tr$r9rrrrrzOpenAI transcription failed: %srXr)rRrr[rOrwrxr?DEFAULT_STT_MODELr$rrrrrrArr_extract_transcript_textrrrrrrrJr.)r(rtrLrrrrrrrrrrFrros r_transcribe_openairs ?AA    XX          ] =[\\\[   @*N_```& [PPPPPPPPPPPP(BTUVVV i&& * & ; B B$#.8K.G.GFFV!C!!                7}EEO KKFi-z3;O;O Q Q Q $?PXYYFGT22E  FGT22E   ``` =^S\=^=^_____ WWW =URS=U=UVVVVVVVV VVV =TQR=T=TUUUUUUUU PPP =N1=N=NOOOOOOOO [[[ 6D III =YVW=Y=YZZZZZZZZ[s ;6;;=FE')+C  E' C$$E''C$(AE';*F'-FFH',H'0F>8H'>H'GH'H'G*$H'* H'7%H"H'"H'c td}|sddddS ddlm}||5}t|d 5}|jj||t|jd  }d d d n #1swxYwYt|}t d t|j|t|d|ddcd d d S#1swxYwYd S#t$r ddd|dcYSt$rB}td|ddddt!|jdcYd }~Sd }~wwxYw)zTranscribe using Mistral Voxtral Transcribe API. Uses the ``mistralai`` Python SDK to call ``/v1/audio/transcriptions``. Requires ``MISTRAL_API_KEY`` environment variable. MISTRAL_API_KEYFrizMISTRAL_API_KEY not setrcr)Mistral)rLr)r& file_name)r1rNz-Transcribed %s via Mistral API (%s, %d chars)Trr9rz Mistral transcription failed: %srXzMistral transcription failed: )rmistralai.clientrrrArcompleterrrrxr?rrrJr.r__name__) r(rtrLrrrrCrFros r_transcribe_mistralrSs[ -..G X =VWWWr,,,,,, WW % % % [i&& *4==$%/d9oo>RSS>                7v>>O KK?Y$j#o2F2F    $?PYZZ [ [ [ [ [ [ [ [ [ [ [ [ [ [ [ [ [ [ ``` =^S\=^=^_____ rrr 7T JJJ =p^bcd^e^e^n=p=pqqqqqqqqrsjC4C'7A?3 C'?B C'B AC' C4'C++C4.C+/C44E E7E EEc "ddlm}|}t|dpd}|sddddSt }|di}t|d p*t d p|d ptd }t|d ptj d pt}t|dd} t|dd} ddl } ddlm} i} |r|| d <| rd| d<| rd| d<t|d5}| |dd|| ddt#|j|fi| d}dddn #1swxYwY|jdkrd} |}|diddp|jdd}n#t,$r|jdd}YnwxYwddd|jd |dS|}|d!d}|sddd"dSt.d#t#|j|d ||d$dt3|d|dd%S#t4$r ddd&|dcYSt,$r0}t.d'|d(ddd)|dcYd}~Sd}~wwxYw)*zTranscribe using xAI Grok STT API. Uses the ``POST /v1/stt`` REST endpoint with multipart/form-data. Supports Inverse Text Normalization, diarization, and word-level timestamps. Requires ``XAI_API_KEY`` environment variable. rrJrLriFzRNo xAI credentials found. Configure xAI OAuth in `hermes model` or set XAI_API_KEYrcrrr2/r0r/rTdiarizeN)hermes_xai_user_agenttruerz/sttzBearer ) Authorizationz User-Agentrx)headersfilesdatarr.messagerzxAI STT API error (HTTP z): r z!xAI STT returned empty transcriptz@Transcribed %s via xAI Grok STT (lang=%s, %.1fs audio, %d chars)rr9rz xAI STT transcription failed: %srXzxAI STT transcription failed: )rPrKr[rIrkrKrr2rstriprrrrrequestsrrpostrr status_coderr rJrxr?rrr.)r(rtrKcredsrLrL xai_configrr0 use_format use_diarizerrrrresponserEerr_bodyrCrFros r_transcribe_xairzs<;;;;; ( ( * *E%))I&&,"--3355G  i   "##Jr**Jz""  + , ,  99Z      eggffSkk  z"" & 90 1 1 & % egg  !$!?!?@@J!*..E"B"BCCK?c888888!  ('D   $#DN  %$DO )T " " j}}!!!%8w%8%8"7"7"9"9 T)__1:>%  H                 3 & &F -#==??!gr2266y"EE\W[X[W[I\ - - -!tt, -! UH4HUUVUU   **VR006688   <   N OO JJz8 , , JJz1 % %      ERRR ``` =^S\=^=^_____ ccc 7T JJJ =a^_=a=abbbbbbbbcs|;1M,AG:. M:G>>MG>MA I! M!I=:M<I==MAMA*MN N%N N Nct|}|r|St}t|sddddSt|}|dkrP|di}t |p|dt }t||S|dkrP|di}t|p|dt }t||S|dkr|pt}t||S|d krC|d i}|p|dt}t||S|d krC|d i}|p|dt}t||S|d kr|pd }t!||St#||} | t%||| ||St'||t(r||ini} | d} |p| d} t+|||| | } | | Sdddt,ddS)a Transcribe an audio file using the configured STT provider. Provider priority: 1. User config (``stt.provider`` in config.yaml) 2. Auto-detect: local faster-whisper (free) > Groq (free tier) > OpenAI (paid) Args: file_path: Absolute path to the audio file to transcribe. model: Override the model. If None, uses config or provider default. Returns: dict with keys: - "success" (bool): Whether transcription succeeded - "transcript" (str): The transcribed text (empty on failure) - "error" (str, optional): Error message if success is False - "provider" (str, optional): Which provider was used Friz4STT is disabled in config.yaml (stt.enabled: false).rcr&r1rrr$rrzgrok-sttN)r*r0rSzZNo STT provider available. Install faster-whisper for free local transcription, configure a or install a local whisper CLI, set GROQ_API_KEY for free Groq Whisper, set MISTRAL_API_KEY for Mistral Voxtral Transcribe, configure xAI OAuth or set XAI_API_KEY for xAI Grok STT, or set VOICE_TOOLS_OPENAI_KEY or OPENAI_API_KEY for the OpenAI Whisper API.)rprKrOrRrIr|rzrr~rrrrrDEFAULT_MISTRAL_STT_MODELrrrrGrrrarj)r(r1r.rLr local_cfgrt openai_cfg mistral_cfgcommand_provider_config plugin_cfgplugin_language plugin_model plugin_results rtranscribe_audiors( ! + +E  "##J * % % K   Z((H7NN7B// +  @Y]]7,?@@  !J777?""NN7B// 3  @Y]]7,?@@  )J???644  :6668^^Hb11 HjnnW6GHH !)Z8889 nnY33 Qkoog7PQQ "9j9995(j y*555C8ZXX*&   #      *2>*b11L>\<_==133N/..2>BBOi % ' '  7< G !!!  *G  ) 0 0 5 5888$-- rPrct|tr|St|dr9t |d}t|tr|St|t r>|d}t|tr|St|S)zBNormalize text and JSON transcription responses to a plain string.r )rr[rkhasattrrrrI)rrs rrrs-%%%""$$$}f%%! v.. eS ! ! !;;== -&&!!!&)) eS ! ! !;;== }   # # % %%rPr)q__doc__loggingrrlr\rr<pathlibrtypingrrr urllib.parserutilsrtools.managed_tool_gatewayr tools.tool_backend_helpersr r r getLoggerrrxrrrrr[rr"rNrO _HAS_MISTRALrMrzrrrrrrjrrWr0rr2rhrrlrvrwrCobject__annotations__rDrrKrOrTr`rdrgrqrsr|r~r frozensetrrr;rrrrrrrrrrrintrrrrr rrr'rGrRrarpry BaseExceptionrzrrtuplerrrrrrrrRrrSrPrrs 6 &&&&&&&&&& !!!!!!CCCCCC  8 $ $ / / / /$IIIIII&o&677oh'' {++ !BI0+>>"#35MNN%BI&9;PQQ24? /+KLL ")13NOO29/1FGGhhh666 MLL ZZZ "& hv%%%#'8C='''$22x~222224%c%hsm%%%%"Xc]""""#hsm#### Xc]    5D5555x}*.x}.....t4" ###<'*##$)!&Y'D'D'DEE8c3h8s8tCH~8888S#X  #s(^<>DcN>t>>>>S#Xd38n* DcN    htCH~.FRV T#s(^     [4S>[c[[[[schsm@#hsmPS(%%sCx.% %%%%P=j.>=4====@(Qc(QE(Qj6Q(Q(Q(Q(QV$##F%) ~~~~ cN~S#X ~ SM ~  #s(^ ~~~~Bldlslllln,0L  " LLLLLc3h(L C= L sm Ld38nLLLLhCHT#s(^,DV  DM Dd D D D DK#KKKK6:a:a#:a$sCx.:a:a:a:azICI3I5#PXY\P]A];^IIII,EaEa#Ea$sCx.EaEaEaEaX-[-[-[c3h-[-[-[-[h2[#2[32[4S>2[2[2[2[rr3rCrDcNrrrrNccsccccS#XccccccccVyyyHSMyT#s(^yyyyxU38_<&C&C&&&&&&rP